Researchers at Kyoto University in Japan have developed an artificial intelligence-based program that can aid in determining the prognosis of idiopathic pulmonary fibrosis (IPF). Their study was published in the Annals of the American Thoracic Society.

The artificial intelligence-based quantitative computed tomography (CT) image analysis software (AIQCT) utilizes high-resolution CT (HRCT) images to classify and quantify 10 types of parenchymal patterns as well as airways. AIQCT then expresses the volumes of airways and each parenchymal pattern as percentages of total lung volume.

The percentages calculated by AIQCT were compared to visual scores from respirologists and radiologists, and they had medium to strong correlations (correlation coefficient, 0.40-0.95) depending on which parenchymal pattern was being compared. Dice similarity coefficients were also computed between the AIQCT and ground truth volumes for reticulation, honeycomb, and bronchi, and they yielded indices of similarity of 0.67, 0.76, and 0.64, respectively.

Continue Reading

“Our newly developed artificial intelligence-based image analysis software successfully quantified parenchymal lesions and airway volumes,” the authors wrote.

Read more about IPF testing

Further analysis of the data and survival outcomes using multivariable Cox regression found that the volumes of bronchi (adjusted hazard ratio [HR], 1.33; 95% CI, 1.16-1.53) and normal lung tissue (adjusted HR, 0.97; 95% CI, 0.94-0.99) were associated with survival outcomes in patients with IPF after adjustment for the Gender-Age-Physiology (GAP) stage. These results implied that “bronchial and normal lung volumes on chest HRCT may provide additional prognostic information on the GAP stage of IPF.”

The AIQCT software was trained using 304 HRCT scans from patients with diffuse lung disease. For software validation, AIQCT results for 30 plain HRCT scans of lung disease were compared to visual scores from a group of 3 pulmonologists and a group of 3 chest radiologists.

The physicians measured the percentages of 9 parenchymal patterns, including normal lungs, bronchi, vessels, reticulation, ground-glass opacities, honeycombing, consolidation, hyperlucency, and nodules in a single axial slice for each patient. The physicians were allowed to look at additional slices to aid in their classification, however. The average scores for the 3 observers in each group were then computed and compared to the scores generated by AIQCT.

AIQCT was validated by comparing detected volumes for reticulation, honeycombing, and bronchi, with areas visually labeled by 1 pulmonologist and 1 radiologist from each group. The volumes were compared for 10 IPF patients and the Dice similarity coefficient was calculated.

The trained and validated AIQCT was then applied to 120 novel IPF HRCT scans and compared to the survival of patients. During the follow-up period (median of 2184 days), 66 patients died and 1 received a lung transplant.


Handa T, Tanizawa K, Oguma T, et al. Novel artificial intelligence-based technology for chest computed tomography analysis of idiopathic pulmonary fibrosis. Ann Am Thorac Soc. Published online August 19, 2021. doi:10.1513/AnnalsATS.202101-044OC